How can representation dimension dominate structurally pruned LLMs?

Open in new window